/RiceSeedlingDataset

A repository to open rice seedling dataset.

Rice Seedling Datasets

A repository to open rice seedling dataset.

Related Publications

The data descriptor was published on Remote Sensing, MDPI. (open access)
A UAV Open Dataset of Rice Paddies for Deep Learning Practice

An application of rice seedling detection using transfer learning was published on Remote Sensing, MDPI. (open access)
Rice Seedling Detection in UAV Images Using Transfer Learning and Machine Learning

MDPI and ACS Style

  1. Yang, M. D.; Tseng, H. H.; Hsu, Y. C.; Yang, C. Y.; Lai, M. H.; Wu, D. H. A UAV Open Dataset of Rice Paddies for Deep Learning Practice. Remote Sens. 2021, 13, 1358. doi:10.3390/rs13071358
  2. Tseng, H. H.; Yang, M. D.; Saminathan, R.; Hsu, Y. C.; Yang, C. Y.; Wu, D. H. Rice Seedling Detection in UAV Images Using Transfer Learning and Machine Learning. Remote Sens. 2022, 14, 2837. doi:10.3390/rs14122837

Contents

  1. Data Download Link
  2. Dataset of Rice Seedling Classification
  3. Dataset of Rice Seedling Detection
  4. CNN Classification Model
  5. Detection Demo Dataset

1. Data Download Link

2018 Orthomosaic image

Filename File size Image size (pixels) Spatial resolution (mm/pixel)
2018-08-07_ARI80_20m_Orthomosaic.tif 465MB 19406 x 10413 5.23
2018-08-14_ARI80_20m_Orthomosaic.tif 610MB 19876 x 10687 5.11
2018-08-23_ARI80_20m_Orthomosaic.tif 557MB 18294 x 9823 5.55

These three links are the orthomosaic image of paddy No.80, TARI, which were in TWD97 / TM2 zone 121 (EPSG:3826) projection. The images were acquired in three consecutive growing stages in 2018.

Expansion Orthomosaic image (2019-2020)

Filename File size Image size (pixels) Spatial resolution (mm/pixel)
2019-03-26_ARI78_20m_Orthomosaic.tif 485MB 20192 x 10858 5.04
2019-04-02_ARI78_20m_Orthomosaic.tif 418MB 19061 x 10250 5.33
2019-08-12_ARI78_20m_Orthomosaic.tif 503MB 21998 x 11829 4.62
2019-08-20_ARI78_20m_Orthomosaic.tif 605MB 21265 x 11435 4.78
2020-03-12_ARI78_40m_Orthomosaic.tif 278MB 15933 x 8568 6.38
2020-03-17_ARI78_40m_Orthomosaic.tif 317MB 15941 x 8572 6.38
2020-03-26_ARI78_40m_Orthomosaic.tif 385MB 15962 x 8583 6.37
2020-08-12_ARI78_40m_Orthomosaic.tif 330MB 15966 x 8586 6.37
2020-08-18_ARI78_40m_Orthomosaic.tif 382MB 15977 x 8591 6.36
2020-08-25_ARI78_40m_Orthomosaic.tif 402MB 15979 x 8593 6.36

These 10 links are the orthomosaic image of paddy No.78, TARI, which were in TWD97 / TM2 zone 121 (EPSG:3826) projection. The images were acquired in 2019 and 2020.

Dataset of Rice Seedling Classification (Train-Val)

RiceSeedlingClassification_2class.tgz (uncompressed size 426MB)

This file includes two folders: riceseedling and arableland. Train-val and test datasets are all included.

Dataset of Rice Seedling Detection (Train-Val)

RiceSeedlingDetection.tgz (uncompressed size 19.1MB)

This is a PASCAL VOC format object-detection dataset, which includes two folders: JPEGImaegs and Annotations.

Dataset of Detection Demonstration (Test)

RiceSeedlingDemo.tgz (uncompressed size 48.5MB)

This file contains 8 detection demo images and the corresponging PASCAL VOC xml format annotations.

An overview of the region of different datasets

An overview of the field no. 80 (cyan bounding area) in TARI, Taichung. Image acquired on August 7, 2018. The green bounding area represents the area for training-validation dataset, and the red bounding area represents the subsets for object detection demonstration dataset.

2. Dataset of Rice Seedling Classification

This dataset contains two classes:

  • rice seedling
  • arable land

An overview of image dataset

The number of images used for training, validation, and testing in the rice seedling dataset.

Class Training Samples Validation Samples Testing Samples Total Samples
Rice Seedling 22,438 561 5,048 28,047
Arable Land 21,265 532 4,784 26,581
Total 43,703 1,093 9,832 54,628

3. Dataset of Rice Seedling Detection

This dataset is used for object-detection model training and validation.

  • PASCAL VOC xml format annotation
  • 3 consecutive mission: Aug 7th, Aug 14th and Aug 23rd
  • 8 subsets
  • 25 samples/subset
  • 600 samples total

Examples of three growth stages of the rice seedling detection dataset

4. A Simple Example of CNN Classification Model

The architecture of the proposed network for rice seedling classification

In the article, we proposed a simple CNN architecture which adopted the stacked convolution layer from VGG-16. The pre-trained model is provided in the model directory.

The environments for the experiments are:

  • host
    • Ubuntu 18.04.5 LTS Server 64bit
    • NVIDIA Display Driver 450.57 (cuda 11.0)
    • Docker CE 20.10.2
    • nvidia-docker2
  • container (image: nvcr.io/nvidia/tensorflow:20.03-tf2-py3 (from Nvidia GPU Cloud))
    • tensorflow 2.2.0
    • python 3.6.9
    • matplotlib 3.3.0
    • scikit-image 0.17.2
    • scikit-learn 0.23.1
    • python3-opencv 3.2.0 (apt install)

To test the provided model, simply call the tf.keras.models.load_model() and you're ready.

5. Detection Demo Dataset

This dataset is used for patch-based object-detection scenario.

  • 8 subsets
  • 8m x 8m region
  • 1527 x 1527 pixels
  • PASCAL VOC xml format annotation

An overview of 8 detection demo images

Subset No. Raw Image Prediction Image Ground Truth Image
1
2
3
4
5
6
7
8

Comparison of Detection Result and Ground Truth

Subset No. 1 2 3 4 5 6 7 8
Prediction 735 1006 1037 809 1004 1050 1017 1032
Ground truth 898 1000 1019 964 971 1002 1033 1005
Error (%) 18.15 0.60 1.77 16.08 3.40 4.79 1.55 2.69